AI-Assisted Documentation vs. AI Medical Scribes: What's Actually Different

There's a lot of loose talk about "AI in healthcare documentation." Two distinct categories get lumped together constantly — and the confusion leads to practices buying the wrong tool for the wrong problem.

AI medical scribes and AI documentation tools both reduce paperwork. That's roughly where the similarity ends. Let's get specific about what each one actually does.

AI Medical Scribes: Capturing What Happened

An AI scribe listens to a clinical encounter in real time and generates a structured note from the conversation. Products like Microsoft DAX Copilot, Abridge, and Nabla sit in the room — or on your phone — and convert the provider-patient conversation into a SOAP note or visit summary.

The core capability is ambient listening. The physician talks to the patient normally. The scribe handles documentation in parallel. By the time the visit ends, a draft note is ready for review.

What scribes are good at:

Capturing clinical encounter details verbatim
Generating SOAP notes, HPI sections, and assessment/plan text
Populating EHR templates directly
Reducing after-hours charting

What scribes are not designed for: generating the downstream compliance documents that payers require. A scribe creates a great clinical note. It doesn't know how to structure a medical necessity letter for a specific payer's biologic formulary criteria.

AI Documentation Tools: Building What Payers Require

This is a different problem entirely. When a patient needs a biologic — say, a TNF inhibitor for ankylosing spondylitis — the prior authorization process requires documentation that is structured around payer-specific criteria, not around how the clinical encounter went.

Payers want to see:

Evidence of step therapy (did the patient fail methotrexate first?)
Diagnosis codes with clinical substantiation
Disease activity scores and lab values in specific formats
Contraindications to alternatives
Language that maps to the payer's Local Coverage Determination (LCD)

A scribe captures what was said. A documentation tool builds what the payer needs to approve the claim. These are genuinely different outputs requiring different inputs.

AI documentation tools for prior auth — like Luma — take structured clinical data and generate compliance-ready letters and medical necessity documentation based on current payer criteria. The input isn't a conversation recording. It's clinical facts: diagnosis codes, lab values, prior treatment history, disease severity scores.

The Cost Comparison

AI scribes from enterprise vendors typically run $300–$600 per physician per month. DAX Copilot, which integrates with Epic, sits at the higher end of that range for full deployments. Abridge and newer entrants are competitive, but pricing scales with seat count.

AI documentation tools for prior auth sit at a different price point and serve a different user — often the billing team or a PA coordinator, not the physician directly. Tools like Luma run around $149/month for a practice, not per seat, because the value is in throughput: how many prior auth documents you can generate per month, not how many physicians are dictating.

The ROI math is also different. Scribes save physician time during and after encounters — measured in charting hours recaptured. Documentation tools save staff time on PA submissions and directly impact first-pass approval rates — measured in fewer denials and faster time-to-treatment.

When You Need a Scribe

Scribes solve a specific problem: physicians spending 90+ minutes per day on charting after clinic hours. If your physicians are doing pajama-time documentation, or if you're losing physician time to EHR data entry during visits, an AI scribe addresses the root cause.

High-volume primary care and outpatient specialties see the most compelling ROI. A primary care physician seeing 25 patients per day, each requiring a visit note, benefits enormously from a tool that cuts note time from 8 minutes to 2.

Scribes also have a measurable effect on patient experience — when the physician isn't looking at a screen typing, the visit feels more human. That's a real benefit even if it's hard to put a dollar value on.

When You Need a Documentation Tool

If your practice's pain is in prior authorization — specifically the documentation burden of generating medical necessity letters for biologics, specialty drugs, or complex procedures — a scribe doesn't solve your problem.

The documentation for a biologic PA isn't derived from the visit transcript. It's assembled from the patient's clinical history across multiple encounters, formatted against payer-specific templates, and submitted through a separate workflow from the EHR note. Ambient listening doesn't produce that output.

Rheumatology, dermatology, gastroenterology, and oncology practices are the highest-need segments here. If you're processing 40+ biologic PAs per month and spending 30–45 minutes per case on documentation, an AI documentation tool has clear and calculable ROI. The AMA's data consistently shows PA burden as one of the top drivers of physician frustration — and the documentation piece is where the hours actually go.

When You Need Both

These tools aren't competitive. A physician might use DAX Copilot to generate their visit note after a rheumatology encounter — and the same practice might use Luma to generate the medical necessity letter for the biologic PA that came out of that visit. The scribe handles the clinical record; the documentation tool handles the compliance output.

The clinical note from the scribe can actually feed the documentation tool. The structured facts captured in the visit note — current disease activity, prior treatment history, lab results discussed — become inputs to the PA documentation. That's the integration story that makes both tools more valuable together.

Some larger health systems are building workflows that connect ambient scribe output directly to PA documentation generation. It's early, but the direction is clear: AI captures the encounter, AI also handles the downstream administrative burden that the encounter triggers.

The One Question That Clarifies Everything

Ask yourself: where is the time actually going?

If the answer is "physicians are spending hours after clinic charting visit notes," you have a scribe problem. Get a scribe.

If the answer is "our staff is spending 40–50 hours per month generating PA documentation for biologics, we're getting one in five denied on first submission, and our physicians are doing peer-to-peer calls that still result in denials," you have a documentation and compliance problem. That's a different tool.

The mistake most practices make is assuming one product solves everything. The market has two distinct tool categories because there are two distinct problems. Match the tool to the actual bottleneck — and run the numbers before you buy anything. More on that math is covered in other posts on the Luma blog.

Sources: American Medical Association, "2023 Prior Authorization Physician Survey" (ama-assn.org); CAQH Index 2023 (caqh.org); Nuance Communications, DAX Copilot product documentation (nuance.com); Abridge product documentation (abridge.com); JAMA Network Open, "Effect of Ambient Clinical Intelligence on Clinician Documentation Time" (2023).